One of the most thought provoking conversations ever. The most mindblowing idea from it: Evolution forms complex multilevel hierarchies. Proteins, Cells, Tissues, Organs, Organisms, Ecosystems, Societies... Each level in the hierarchy has it's own set of goals, which are controlled from the level higher up, for the purpose of serving that higher level.
Bodies control what organs do, yet the organs have autonomy. Organs control what tissues do. Yet the tissues have autonomy. Tissues control what cells do. Yet those have autonomy to. Xenobots are what happens to otherwise normal frog cells if you remove them from that high level control.
Zooming into anything leads to physics and chemistry, not fairy dust.
Arranging physics in the right way leads to emergent things, selfs that can have goals, preferences, anticipations, that are not explained by the physics alone.
Where do Selfs come from? Selfs and Cognition are a fluent smooth continuoum. All the way from unfertilized cell to complete human the self gradually builds.
Evolution changes not the cells themselves, but the signals between cells to make something new. It changes the "software". There is a continuum between the two again.
Multi Hierarchy Competency Architecture. Everything in life has goals. DNA, Proteins, Cells, Tissues, Organs, Organisms, Colonies.
Who is responsbile for that Hierarchy? Why does evolution build structures like that? How is that built?
Engineering is the right perspective. Asking how to build it leads to quantifying and clarifying what you mean and therefore helps understanding and breaking down of complex topics.
Biology – competition within and across levels produces final outcomes
Body parts in developing organisms can "find their way" to where they belong, and we are just beginning to understand how cells know where to go to form organs and body parts.
Parts of living organisms have competency and goals to achieve and attain. Among each level of the hierarchy the goals and competences change and they don't care about what's going on in the levels way below or way above them!
The world that built us is the one we reject in some sense when we construct human societies. – Lex Fridman
Bending the option space to change what outcomes cells produce.
Hierarchical systems accomplish goals at all levels, by doing and distributing computation along all levels.
Every system has some degree of intelligence in it's own problem domain. – Michael Levin
Cells solve problems in physiological and transcriptional space. Collectives of cells solve problems in anatomical space. Bodies solve problems in social spaces, linguistic spaces, 3d spaces. Groups of bodies/individuals solve problems in financial and societal spaces.
This is "myself" because this is stuff that I have more control over than all the other stuff. The idea of self literally can grow and shrink depending on that definition.
The world has agency and turning that onto the part of the world that is "myself" I have agency. That's what free will is. Agency to act within the world to accomplish goals + the insight that because there is only limited energy + time that "I" as an agent can make use of, I have to make choices as to what things to do and what things not to do.
Any agent that self-constructs under energy constraints is going to believe in free will. – Michael Levin.
The sense of agency is really useful to accomplish tasks, simple or complicated. – Lex Fridman
Facial Symmetry is a judgement of beauty because it is correlated to how good your cells are at coordinating shape and keeping shape together. It's a "half-assed" way of judging the state of somebodies biochemistry by their looks.
Arms races between parasites and biological systems leads to interfacibility but not too much interfacibility, because that would mean hijacking would be too easy.
Is there some evolutionary value to beauty itself? – Lex Friedman
How does a collection of cells remember something? Where do Neurons come from? Ancient Networks of cells to control amorphous space (i.e. body shape) came before neurons. Ion Channels, Electrical Synapses and the other stuff that neurons use is older than multicellular life!
Electrical Networks are good at having memories, i.e. at integrating information across space and time, and thereby can do optimization tasks (image recognition etc.)
What is bioelectricity? Single Cells with Membranes allow ions to go through or not, building up charges and voltage gradients across membranes.
Gap Junctions and Consciousness.
Ion Channels and Gap Junctions are voltage gated/sensitive conducting elements => in other words => transistors. Hence turing completeness is there.
Gap junctions, to some extent, wipe ownership information on data. – Michael Levin
Cells at gap junctions do not "know" where information originated, whether it came from the other cell or from itself. Hence "memory" between these cells as an electrical concept/signal merges. "Mindmeld". Gap Junctions are the beginning of an aggregation of collective intelligence.
Where is the state stored? Many layers. => Biochemical Networks. Genetic Switches as an example. => Cytoskeletal Structures. => Electrical "Flip Flop" Switches, inside of cells. Cycling of current.
There is an equivalent to "volatile" RAM inside of cells.
Hardware change doesn't change, the dynamics of the system govern the state of the information contained within it. If looking at an X-Ray you wouldn't see information contained in a flip flop.
Mappings between computer hardware and state storage and different strategies employed within cells works reasonably well.
Biology is all about making really important decisions really quickly on very limited information. That's what biology is all about. You have to act, you have to act now. The stakes are very high. And you don't know most of what you need to know to be perfect. – Michael Levin
Instead active inference, efficiency, optimization over time.
I am going to program this computer by changing the melting point of copper. Maybe you can do things that way... but my god it's hard to program at that level. – Michael Levin
Doing things at the wrong level of detail is a recipe for desaster and unintended side effects, but that's exactly what medicine does right now.
We don't program with a soldering iron. – Micheal Levin
Focus on the "goal" states of – this is what a healthy body should look like – will usher in a new paradigm in medicine: Somatic Psychiatry. Soma is the body. Somatic cells are body cells. They form electrical networks, these networks interact.
Planaria hold the answer to every deep question about life. – Michael Levin
Planaria live forever, some of them are 400 million years old in the lab of Michael Levin in physical continuity. Planaria can grow a full body from only a part. They don't have cancer. They can learn. Regrown planaria from the tail parts can remember things from the original brain, even though their brain is regrown from scratch.
Humans (female) can grow a full body from a single cell.
How do Planaria fragments know if they need to grow a "head" or a "tail"? Electrophysical cell network is remembering the shape, mess up the pattern and the planaria regrow in a different way. This network is handed down when worms split up and separate in half to replicate. No genome changes, yet, persistent physiological changes.
Hardware of biology implements software in electronic networks between cells. Those networks can be changed and "programmed" to make different patterns appear, without changes about the hardware (the genes of the cells)
Planaria genomes look like tumors, because the accumulate mutations over millions of years. Yet their bodies are perfeclty shaped and can even regenerate. Genomes don't necessarily determine body shape.
How does genome interplay with cellular electrical networks and shape memory? If competency of cells is good enough, then selection can't pick out "good" vs. "bad" genomes, because all of them produce solid bodies that reproduce somehow...
Biology optimizes for problem solving instead of memory and recall. It doesn't pay a lot of attention to the past and optimizes to make things "work" no matter the circumstances. Because that's what get's selected for. Resiliency.
Binary distinction between robots and living things "organisms" will blur more and more in the future, until they become absolutely meangingless.
Robotics is the rational construction of machines that have useful purposes. – Michael Levin
The substrate of a robot can be frog cells. Xenobots are made from Xenopus laevis cells. African clawed frog. Xenobots are not specific to frogs, they can be made from most stem cells.
Development of an embryo into a human is a regeneration event.
Newts developing kidney tubules. When cells are too big, because messed with by humans, instead of multiple cells forming the tube by cohesion and being glued together, a single gigantic cell stretches around onto itself and changes it's cytoskeleton to form the tube.
The ability to get to the same goal by different means is the hallmark of intelligence. – William James
Cells are intelligent, because they "get their job done" even when circumstance change drastically.
The environment forces the development of cells into the "correct" shape of the organism. Overall the shape is adaptive somehow, no matter the environment, as xenobots show.
Michael: The human brain allows us to do things that we couldn't do without it. Lex: You could say the same thing about the liver. And the heart. Smiles. One of my most favorite parts:
Other things besides the human brain might have meaningful levels of cognition.
Building synthetic biology is necessary to understand biology because otherwise the sample size is just N=1 and we can't make conclusions about biology in general from our limited point of view.
If anything among life "out there in the universe" is universal and fundamental, it's resource limitation, drawing an arbitrary boundary between self and other, minimizing surprise, coarse grained experience.
Selfs are defined by the goals an agent could pursue.
Cognition in Plants is not weird compared to any kind of life we might find out there. Exobiology might be truly weird.
Some of the "magic" of cognition is lost when it is fully designed.
Outcome can be unexpected even when the input is known. Example - the Mandelbrot Set.
Unconvential Cognition. How do you recognize cognition outside of brains and neurons? What other implementations are there for minds? Cellular Automata. How do you have a conversation with unconvential "minds"?
Gene Regulatory Networks can have associative memory. Habitization. Memorization.
If you can train something, you can communicate with it. "Good" actions vs. "bad" actions.
Does a cellular automata "talk" back and change the researcher researching them? What about lab rats? Or organisms that we "domesticated"? Cats are "running" the world.
Cognition is not consciousness. Sentience is something different yet. All of them have different definitions, depending on the person you talk to. The important idea is the engineering parts about them.
Is the feeling of consciousness, the feeling of being a self, different from the biological machinery that creates it? Is there something more to us or is that feeling of "being" simply a side product of biological neural networks?
Grounding problems in Engineering is helpful in researching them and actually finding out how they work. Grounding consciousness in that way is pretty hard or maybe impossible, at least for now. Anil Seth still tries it though.
What format would consciousness theory make predictions in?
If it looks like it is intelligent, it is intelligent. – Lex Fridman
Things that appear conscious, are probably conscious as well.
Finding the other even though there are fundamental similarities is something that humans excel at. Grouping people by the slightest of differences into "us" and "them" to exploit and harm the "other".
The actions you take not only change your payoff, but they change who or what you are. And you could take an action after which you don't exist anymore. – Michael Levin
Where did evolution come from? Evolution is inevitable. Heredity, Imperfect Heredity and Competition => Evolution. Pretty much anything can evolve.
If you re-ran the world a million times, how many times would you get Hitler?
Are there "deep attractors" on Earth that act on evolution, getting things to the same place with a higher than average likelyhood?
Is the competency hierarchy a self correcting system, "destined" to do something? Is life headed towards something?
Predicting the cognitive goals of composite systems. Answering the question of if I have a collective of agents, what form of intelligence does it have? And what kind of goals does it form? Existential question, because we are building these kinds of systems all the time.
Evolution is optimizing for biomass. It's not optimizing for your happiness. – Michael Levin
Natural is not a good compass for moral good.
Building an anatomical compiler. Define what kind of shape you want to have. Get back stimuli. And then make cells grow it, by applying the stimuli.
Ethical norms should change, and will eventually change, to take into account the information processing capabilities, the cognition, of cells and generalizes over all kinds of different life forms.
Can I eat it? Or can I have sex with it? The two fundamental questions of the human condition. – Lex Fridman
How does regenerative medicine work today? 3D printing. Scaffolds, fill them with cells. Stem cells. But more complex structures like eyes can't be regenerated that way yet.
Frog legs that can be regrown by giving them a 24 hour stimulus that essentialy says "hey here was a leg, please regrow it"
Why is there anything but cancer? When cell disconnects from electrical network that informs about high level goals, they revert to uni-cellular lifestyle. That's cancer. The disconnect changes the level of self-hood from the orgism to the cell only. Goals of single celled organisms are proliferation and migration. This is what cancers do. Gap Junction closes, connection is severed, the rest of the body becomes environment, food to be consumed.
Bioelectric state is more powerful than genetic mutation. Mutation that says tumor can be suppressed when bioelectric state is kept up.
Viruses, on their own, don't have minds. Because they don't do anything, they just "sit there". But they can interact with "other minds", cells, and influence them.
Both virus and cell are transformed by the experience, and in that sense, both are living. – Lex Fridman
Unask the question of "is it alive" and replace it with – what is its level of cognition?
All cognitive claims are engineering claims.
Terms coined by Michael Levin: Agential Material Teleophobia Anatomical compiler Cognitive Lightcone (outer bound of what a systems outmost goals can theoretically be, spatial and temporal goals you can pursue and achieve, the more complex the system, the bigger the lightcone, it's a way of thinking about how much capability a system has, "how big are the goals you are capable of envisioning and working towards") Morphoceutical Ionoceutical
Michael Levin came from computer engineering.
It's dangerous to give advice because things change so fast. – Michael Levin
Pay attention to speficic critique and feedback, "here, this is wrong because x,y,z" instead of "generally something like this is what you should do".
Read very broadly, work really hard, know what you are talking about, take all specific criticism as an opportunity to improve what you are doing and then completely ignore everything else. – Michael Levin
At best we know what we should be doing, we very rarely what anybody else should be doing. – Michael Levin
Criticism from outside is often not useful data. It could be that what you are doing is something radically new and superior and you would still get attacked for it heavily.
Building up your own intuition instead, is key.
Life is hard, and science is hard. – Michael Levin
What's the role of death in life? It promotes change and turnover. What was the first thing to die? The boundary is tricky to think about and not exactly defined. If something dies, most of the cells are still alive. Why don't they go away and live another life as something else?
Human Cognitive Lightcone is bigger than human lifetime. We can think about and work towards goals that are not achievable for us. If cognitive light cone is small enough, all goals are achievable. For humans, that's completely not the case.
Mammals sometimes "give up and die". Why? What is the simplest system that is able to do suicide? Why should a life form discount the idea of surviving at all costs?
Population Level selection is very controversial. Problem is that if you do something where you die, without propagating, this would mean your genome died out.
What's a good way of dying? There is nothing "heroic" you could do. You can't "sacrifice" yourself. Death is not pretty.
Personal experience feels like you are immortal. You get lost in living.
If you don't believe in free will, you still have to pick something to eat.
For practical purposes, finiteness of life, doesn't make much of a difference. You still make plans and think about the future, even though you could be hit by a bus tomorrow.
What's the meaning of life? Not a well posed question.
Whatever the truth is, none changes the fact of how consciousness feels like. The material or non-material that this is made out of in the end, doesn't matter. Soul, neural network, matrix, biological machine, a bunch of cogs. Whatever.
Experience is the thing that matters from a first person perspective.
For science, these questions matter, for anything else not really. How to go from "is" to "should" is not easily possible. Deriving morality from facts makes not much sense.
Science on consciousness will alter what you are, and how you are.
Science and nothing else, is about the process. Discovering what is really true. Understanding how things work, better and better. It's the rational investigation of ourselves and our surroundings.